User recommendation method integrating hierarchical graph attention network with multimodal knowledge graph.

Area of Science:

• Artificial Intelligence
• Computer Science

Background:

• Graph neural networks (GNNs) leverage user interactions but often miss item semantics and multimodal features, limiting recommendation diversity and accuracy.
• Existing methods struggle with cold-start scenarios and integrating diverse data types for user and item representation.

Purpose of the Study:

• To enhance user and item feature representations by integrating knowledge graphs, GNNs, and multimodal information.
• To address limitations in recommendation diversity, accuracy, and the cold-start problem.

Main Methods:

• A novel recommendation model integrating a hierarchical graph attention network with a multimodal knowledge graph.
• The model incorporates collaborative knowledge graph neural layers, image feature extraction, and text feature extraction.

Main Results:

• The proposed model significantly outperforms existing recommendation methods on two public datasets.
• Demonstrated improvements in recommendation performance through enhanced user and item feature extraction.

Conclusions:

• The integration of knowledge graphs and multimodal features with GNNs offers a powerful approach to enhance recommendation systems.
• The model effectively addresses limitations of traditional GNNs, improving accuracy and handling new users/items.